Staged Training Report ✓ Complete

Run ID: stage10batch8run
Generated: 2026-02-10 22:34:26
Stages Completed: 1
Total Elapsed Time: 00:37:04

Configuration

Config Defaults Changed Since Last Commit

Parameter	Previous	Current
`batch_size`	1	8
`runs_per_stage`	1	2
`serial_runs`	None	True

All Configuration Parameters (59 parameters)

Parameter	Value
`total_samples`	10000000
`batch_size`	8
`stage_samples_multiplier`	100000000000
`update_interval`	250
`window_size`	100
`num_best_models_to_keep`	1
`sampling_mode`	Loss-weighted
`loss_weight_temperature`	0.5
`loss_weight_refresh_interval`	50
`stop_on_divergence`	True
`divergence_gap`	0.002
`divergence_ratio`	1.3
`divergence_patience`	10
`divergence_min_updates`	5
`val_spike_threshold`	2.0
`val_spike_window`	15
`val_spike_frequency`	0.75
`val_plateau_patience`	100
`val_plateau_min_delta`	0.0001
`custom_lr`	0.0001
`disable_lr_scaling`	True
`custom_warmup`	-1
`lr_min_ratio`	0.001
`resume_warmup_ratio`	0.05
`plateau_factor`	0.8
`plateau_patience`	20
`preserve_optimizer`	False
`preserve_scheduler`	True
`samples_mode`	Train additional samples
`num_random_obs_to_visualize`	2
`selected_frame_offset`	3
`runs_per_stage`	2
`serial_runs`	True
`clean_old_checkpoints`	True
`enable_baseline`	False
`baseline_runs_per_stage`	1
`run_id`	stage10batch8run
`enable_wandb`	True
`wandb_project`	developmental-robot-movement
`lr_sweep.lr_min`	1e-07
`lr_sweep.lr_max`	0.01
`lr_sweep.phase_a_num_candidates`	5
`lr_sweep.phase_a_seeds`	1
`lr_sweep.phase_a_time_budget_min`	3.0
`lr_sweep.phase_a_survivor_count`	2
`lr_sweep.phase_b_seeds`	3
`lr_sweep.phase_b_time_budget_min`	10.0
`lr_sweep.ranking_metric`	median_best_val
`lr_sweep.min_samples_before_timeout`	1000
`lr_sweep.min_evals_before_stop`	5
`lr_sweep.save_sweep_state`	True
`plateau_sweep.enabled`	True
`plateau_sweep.plateau_ema_alpha`	0.9
`plateau_sweep.plateau_improvement_threshold`	0.0005
`plateau_sweep.plateau_patience`	100
`plateau_sweep.cooldown_updates`	5
`plateau_sweep.max_sweeps_per_stage`	2
`plateau_sweep.min_sweep_improvement`	0.0
`stage_time_budget_min`	180

Timing Summary

Stage	Plateau Sweeps	Sweep Time	Training Time	Stage Total
Stage 10	0	00:00:00	00:11:27	00:11:27
TOTAL	0	00:00:00	00:11:27	00:11:27

Stage Results

Stage	Best Loss	Stop Reason	Samples Trained	Time	Sweeps	LR (Initial→Final)
Stage 10	0.022091	divergence	83,712	00:11:27	0	1.0e-04

Total Plateau Sweeps: 0

Stop Reason Breakdown

divergence: 1 stages

Loss Across Full Training Run

Loss Detail (Post Initial Drop)

Multi-Run Statistics

Total Runs: 2
Average Best Loss: 0.024545 ± 0.002453
Best Overall: 0.022091
Worst Overall: 0.026998

Stage 10 (2 runs)

Run	Best Loss	Stop Reason	Samples	Time	Selected
1	0.022091	divergence	83,712	00:11:27	✓
2	0.026998	divergence	12,544	00:25:03
Mean: 0.024545 ± 0.002453		Min: 0.022091 / Max: 0.026998		Range: 0.004907

Best Checkpoint

Name: best_model_auto_session_so101_should_pan_500_stage10_train_304_stage10batch8run_00080384_fresh_val_0.022091.pth
Stage: 10
Hybrid Loss (full session): 0.022522

Learning Rate Timeline with Plateau Sweeps

Stage Progression

Stage	Orig Loss	Train Loss	Time	Samples	Stop Reason
10 ⭐	0.022522	0.022091	00:11:27	83712	divergence

Hybrid Loss Over Original Session (per Stage)

Stage 10 (Best) - Hybrid Loss: 0.022522

Sample Counts

Cumulative Across All Stages

Per Stage

Stage 10 (Best) - Total Samples: 83,712

Best Checkpoint Inference

Selected Frame 3

Action 0

Action 1

Action 2

Random Observations

Observation 207

Action 0

Action 1

Action 2

Observation 302

Action 0

Action 1

Action 2

Individual Stage Reports

Stage 10 Run 1 (Staged) (orig loss: 0.022522) ⭐